An agent does not read your marketing page: it calls your endpoint. Agent Readiness is the panel that runs inside your connected store and checks what you announce against what you actually respond, alongside what a neutral third party says and the agent traffic we have really observed.
Today it is only visible to an already-connected store — there is no preview without connecting anything, and we say that plainly.
A person who sees a wrong price asks or leaves. An agent acting on instructions may try to buy at that price, hit a dead endpoint, or give up without you ever knowing. The drift between what you declare and what you serve is invisible until someone — or something — runs into it.
search_products, get_product_details, create_cart: if an advertised tool does not respond as advertised, the agent does not improvise a workaround the way a person would.
What you declare has to be what you actually serve.A catalogue price that no longer matches what gets charged, a returns policy that says different things depending on where it is read: none of this breaks your store for a human, but it does for whoever compares automatically.
Catch it before an agent does, in production.How many agents have arrived, where they fall off, how many searches return nothing: it is what has actually passed through your endpoint, not an estimate.
Your own data, not a market forecast.No probe is a bare yes or no. Each one says what it looks at, what it found and, if it could not run, why — an unexplained gap reads as a problem even when it is not one.
Agent Readiness exists to catch self-deception, including our own: when a probe has to pick between two readings, it picks the one that does not flatter.
Findings are ranked by really measured impact — abandoned carts, failed calls — before theoretical severity.
If a probe needs a connector or order history and your store does not have it yet, the panel says so — it does not disguise it as a failed check.
This page describes the exact mechanism: the 16 probes, the three columns, and why they are never combined into one score.
They are never averaged or combined. The panel's own data contract has no field to hold a joint score — it is the only lasting defence against grading your own exam.
The panel does not end in a figure: it ends in the list of what to look at first.
From the real catalogue (`probe-catalog.ts`), the source of truth a test ties against the probes that actually run: none can be declared that does not exist, and none can exist that is not declared.
3 probes (C3, C4, C5) need your store connected — they compare against your real catalogue. 1 (C16) needs delivered order history. The rest run without extra credentials. None are offered today without connecting the store.
Without Agent Readiness, the drift between what is declared and what is served surfaces only after it has already cost a sale or a complaint.
No. Trust Score answers whether an agent should trust your merchant; Agent Readiness answers whether an agent can actually operate there. They are two axes, shown together and never added up.
Not today. The panel needs your store connected because it compares against your real catalogue and traffic — without that there is nothing to check against. This page describes exactly what you will see once connected.
The reason is stated: the connector is missing, order history is missing, or there simply was no data that time. An unexplained gap would read as a failure, and it is not one.
No, deliberately. A third party can say one thing and your real traffic another — that is already information — and averaging them would force us to decide which one is right, which is not our call to make.
Nothing described here is a mock-up: it is the 16 probes and three columns running today against connected stores.
